MaxEnt-Based Potential Distribution Mapping and Range Shift under Future Climatic Scenarios for an Alpine Bamboo Thamnocalamus spathiflorus in Northwestern Himalayas

Abstract

Thamnocalamus spathiflorus is a shrubby woody bamboo invigorating at the alpine and sub-alpine region of the northwestern Himalayas. The present investigation was conducted to map the potential distribution of Th. spathiflorus in the western Himalayas for current and future climate scenario using Ecological Niche Modelling (ENM). In total, 125 geo-coordinates were collected for the species presence from Himachal Pradesh (HP) and Uttarakhand (UK) states of India and modelled to predict the current distribution using the Maximum Entropy (MaxEnt) model, along with 13 bioclimatic variables selected after multi-collinearity test. Model output was supported with a significant value of the Area Under the “Receiver Operating Characteristics” Curve (AUC = 0.975 ± 0.019), and other confusion matrix-derived accuracy measures. The variables, namely precipitation seasonality (Bio 15), precipitation (Prec), annual temperature range (Bio 7), and altitude (Alt) showed highest level of percentage contribution (72.2%) and permutation importance (60.9%) in predicting the habitat suitability of Th. spathiflorus. The actual (1 km² buffer zone) and predicted estimates of species cover were ~136 km² and ~982 km², respectively. The predicted range was extended from Chamba (HP) in the north to Pithoragarh (UK) in southeast, which further protracted to Nepal. Furthermore, the distribution modelling under future climate change scenarios (RCP 8.5) for year 2050 and 2070 showed an eastern centroidal shift with slight decline of the species area by ~16 km² and ~46 km², respectively. This investigation employed the Model for Interdisciplinary Research on Climate (MIROC6)–shared socio-economics pathways (SSP245) for cross-validation purposes. The model was used to determine the habitat suitability and potential distribution of Th. spathiflorus in relation to the current distribution and RCP 8.5 future scenarios for the years 2021–2040 and 2061–2080, respectively. It showed a significant decline in the distribution area of the species between year 2030 and 2070. Overall, this is the pioneer study revealing the eco-distribution prediction modelling of this important high-altitude bamboo species.

1. Introduction

Bamboos support an international trade of over USD ~2 billion per year, with the domestic usage estimated at least 80 per cent of the total, thus forming a major world commodity [1]. Evolutionarily, in the lowland tropics of Gondwanaland, woody bamboos progressed as an ancient group of forest plants during the Tertiary period of the Cenozoic Era [2]. It signifies the importance of bamboos in relation to socioeconomic, spatiotemporal, and spatial distribution across the globe. Despite the substantial utilization and potential, shrubby bamboo in India has been poorly investigated due to inaccessibility and difficult mountain terrain, and knowledge of its diversity and distribution is lacking. However, indiscriminate extraction, developmental activities, and growing forest-fire incidences cause rapid loss of biodiversity, particularly the understory faunal diversity [3]. For instance, the African bamboo species Yushania alpina provides shelter and food to the endangered Tragelaphus euryceros ssp. isaaci (eastern or mountain bongo) in the Aberdare Mountains in Kenya [4]. Another close association of bamboo with an endangered [5] mammal Gorilla beringei beringei (mountain gorilla) occurs in the eastern Democratic Republic of the Congo, Rwanda, and southwestern Uganda [6]. As per the IUCN and the Indian Wildlife (Protection) Act (1972), Ursus arctos isabellinus (Himalayan brown bear) thrives well in the bamboo forest of the Indian subcontinent [6,7]; (http://www.bearconservation.org.uk/, accessed on 13 February 2024). Similarly, one of the most endangered and rare reptiles in the world, Geochelone yniphora (angonoka or ploughshare tortoise), resides in the bamboo-based resources of Madagascar [8]. All these studies highlighted that mountain bamboo species play a dual role in uplifting the socioeconomy and conserving a wide array of wildlife habitats. In view of this, it is of utmost importance to conserve and manage such genetic resources supporting the wide biological diversity.

One such temperate woody bamboo species, Thamnocalamus spathiflorus (Tham ringal), naturally grows widely in the subalpine to alpine zone of the Himalayas. It is not only ecologically important but also has socioeconomic significance for the mountain dwelling community and local artisans [9]. However, a lack of adequate scientific knowledge in most high-altitude bamboos of India is a major impediment in their conservation and management. The fundamental knowledge required to bridge the information gap includes the pattern of distribution, estimation, and mapping of area cover, ecological association, future prediction under a climate change scenario, species richness, taxonomic description, genomic resources, and so forth. Ecological Niche Modelling (ENM) is one of the preferentially used approaches for identifying, mapping, forecasting distributions, and determining change shifts for a vast number of organisms [10,11]. ENM predicts habitat suitability through the mathematical modelling of spatial occurrence records with environmental covariates, via correlative or mechanistic approaches [12,13,14].

Besides the socioeconomic significance, Th. spathiflorus is an ecologically important bamboo taxon occupying a top of the altitudinal range of bamboo and is the first to experience the changes induced by climatic and anthropologic factors. Thus, it is an ideal bamboo species to be investigated for distinguishing the environmental variables associated with its habitat and predicting a range shift in response to future climate projections. In this view, the present study aimed to (1) determine the key contributory bioclimatic variables in the habitat suitability range prediction; (2) map and quantify the area coverage under the species as per the present climatic scenario; and finally (3) estimate the shift under projections of future climate change and overlay the probabilistic distribution using the Köppen–Geiger Climatic Classification (KGCC) system. Overall, a biogeographical habitat and species ecology has been demonstrated in broader perspective of northwestern Himalayas for fixing the conservation priority of this valuable genetic resource.

2. Material and Methods

2.1. Collection of Records of Species Occurrence, Bioclimatic Layers, and Distribution Modeling

In preparation for the field survey, we reviewed the records of species occurrence from various sources, such as the status map of the Forest Survey of India (FSI 2019) and the working plans of the forest departments in the northern Indian states of Himachal Pradesh (HP) and Uttarakhand (UK). We also acquired data on biodiversity and distribution from documents available in the National Forest Library Information Centre (NFLIC) of the ICFRE-Forest Research Institute (FRI); the Northern Regional Botanical Survey of India, Dehradun; and the ICFRE—Himalayan Forest Research Institute (HFRI), Shimla. Furthermore, we gathered information about the actual extent of distribution in a specific forest area from the ground staff of the forest department and forest-dwelling communities.

From 2017 to 2021, we conducted extensive surveys in the states of HP and the UK. We implemented multi-phase random sampling using zig-zag 100 m linear transects in an unbiased manner, with a particular emphasis on the hill slopes where mountain depressions occur. Based on the number of clumps in a linear transect of 100 m, surveyed sites were classified into categories, viz., (i) reduced (5–10 clumps), (ii) disturbed (11–20 clumps), (iii) fair (21–50 clumps), (iv) healthy (51–75 clumps), and (v) pristine (>75 clumps). Additionally, we classified the population distribution as either random, indicating uneven culm distribution in mixed forests, or uniform, indicating gregariously and evenly developing clumps. Furthermore, the phytocoenological analysis also collected data on the primary associate tree species that allow for Th. spathiflorus to grow under their canopy. We employed a Global Positioning System (GPS; Garmin, Olathe, KS, USA) to record geospatial parameters for the dispersed distribution of Th. spathiflorus. We delineated the Th. spathiflorus sections in polygons using an area demarcation tool in the GPS to ensure the accuracy of the sampling [15]. In addition, we utilized the geo-coordinates and GPS-generated polygons to delineate the locations of individual regions and populations, subsequently converting them into a point shapefile. For the purpose of visible interpretation and demarcation, we converted these into a high-resolution base map in ArcGIS. We rectified the sampling bias by removing 156 points from the same grid and incorporating the remaining geospatial data, which were either 800 m linearly apart or 0.64 km² spatially apart [16].

Prediction modeling was conducted using 19 bioclimatic (30 s~1 km² spatial resolution; WorldClim Version 2.0) and 6 climatic variables, 3 spatial analysts, direct normal irradiance (DNI), pedologic, and 6 climatic variables, in addition to occurrence data. The monthly temperature and rainfall values, which are accessible on the WorldClim website (https://www.worldclim.org/, accessed on 15 February 2024), were used to generate a raster layer that represented a variety of bioclimatic data, including seasonality, extreme environmental factors, and annual trends. A Digital Elevation Model (DEM) with a 30 m resolution of the Cartosat-1 from the Indian Geo-Platform was used to generate altitudinal gradients, aspect, and slope [17]. The DNI was downloaded from Solar Energy Centre, Ministry of New and Renewable Energy (MNRE), Government of India (GoI), New Delhi (https://maps.nrel.gov, accessed on 23 February 2024). Additionally, we obtained soil data from the India Dataset of the Soil and Water Assessment Tool (SWAT; https://swat.tamu.edu/data/india-dataset, accessed on 18 February 2024). The present prediction was mapped using data from the 1970–2000 period. In accordance with the 5th Assessment Report (AR5) of the Intergovernmental Panel on Climate Change (IPCC), we examined the future distribution scenarios for one (Representative Concentration Pathway, RCP 8.5) out of four GHG trajectories for 2050 and 2070. On the other hand, in order to improve the accuracy of future species prediction, we employed the shared socioeconomic pathways (SSPs) Model for Interdisciplinary Research on Climate (MIROC6)-SSP245 to estimate the probability distributions of future species for two distinct time periods, 2021–2040 and 2061–2080, for Th. Spathiflorus in both the Himalayan states. Therefore, we used the software ArcGIS 10.8 and ERDAS Imagine 2020 to convert all types of climatic data from various sources into digital number (DN) values. The Pearson correlation coefficient I was then determined using SPSS software (Ver. 16.0) to find the cross-correlation (≥±0.80) between variables (Supplementary Table S1). Ultimately, we generated the probabilistic distribution of Th. spathiflorus by selecting 13 variables and running them in the MaxEnt model.

2.2. MaxEnt Modelling for Current and Future Projection

The open-access program MaxEnt ver. 3.1 is utilized to execute the MaxEnt prediction, which represents the vastly possible number of variables in a space [18]. In this case, the response to the environmental conditions was determined by linear, quadratic, and hinge features [19]. On the other hand, parameters like the maximum number of background points (10,000), categorical (0.250), and threshold (1.000) achieved the model’s most optimal output. The regularization multiplier value of 0.1 was used to mitigate overfitting and overprediction in the model [20] with 5000 iterations [21]. We used about 70% of the geo-coordinates for training of the model and the remaining 30% points were used for validation.

The final probability distribution map (model output) was generated by averaging the predictions from 100 models or replications. Ground control points based on XY coordinates and demarcated boundaries were employed to conceal overpredicted regions. Further, the MaxEnt model yields AUC, which ranged from 0 to 1 [22,23,24]. In addition, the confusion matrix was employed to derive other metrics, including the kappa coefficient (K), Normalized Mutual Information (NMI) n(s), and True Skill Statistic (TSS), to evaluate accuracy [25,26]. Herein, the presence and pseudo-absence points’ single shapefile, with a field value of 1 for the presence and 2 for pseudo-absence, were used in ArcGIS. In the next step, MaxEnt output was re-classified into two classes, viz., class 1 for presence and class 2 for pseudo-absence. Using the “extract value to point” option, a raster value was extracted from a previously generated shapefile. Afterwards, the frequency tool was used to generate a frequency table, showing a frequency of shapefile fields’ value. Finally, a pivot table was created using a frequency table, which showed the upper left value as true presence (a), the upper right value as true false (b), lower left as false true (c), and lower right as the true absence (d). Finally, the precision measures were calculated using below mentioned formulas.

Sensitivity is a/(a + c)

Specificity is d/(b + d)

TSS is sensitivity + specifici–y − 1

Kappa is [(a + –) − (((a + c) (a + b) + (b + d) (c + d))/N)]/[–N − (((a + c) (a + b) + (b + d) (c + d))/N)]

NMI(s) is [−a × ln(a) – b × ln(–) − c × ln(c) – d × ln(d) + (a + b) ×ln(a + b) + (c + d) ×ln(c + d)]/[n × ln(–) − ((a + c) × ln(a + c) + (b + d) × ln(b + d))]

The relevance of the bioclimatic variables was evaluated using response curves and jackknife tests [20]. The response curves quantitatively determine the logistic probability of the species’ presence in its natural distribution range. While holding all other environmental factors at their average sample value, the curves demonstrate how changing an individual variable impacts the projected likelihood of presence. On the other hand, the red curves represent the average answer from all 20 MaxEnt runs, and the blue ones represent the average plus or minus one standard deviation (two shades for categorical variables). We first rendered the final maps using the “degree” unit in the World Geodetic System 1984 (WGS-1984) projection. After that, we used ArcGIS (Projection and Transformation Tool) to convert it to a meter and re-project it into the Universal Transverse Mercator (UTS) system using zones 43 (HP) and 44 (UK). Once the transformation was completed, we used a raster calculator to calculate the area under the prediction output. At last, we completed the development of eco-distribution maps and area calculations for the current and potential distribution scenarios. Notably, the current distribution map was again overlaid on the KGCC system (1976–2000) [27] (http://koeppen-geiger.vu-wien.ac.at/shifts.htm, accessed on 9 February 2024) and Sentinel (Supplementary Table S2) to unravel climatic and spatial attributes information.

3. Results

3.1. MaxEnt Performance and Variables Contribution

The model output was validated using a 30% test dataset with a prediction threshold of 0.7, revealing excellent prediction with 90% points correctly overlaid on the predicted distribution area. A high value of AUC (0.975 ± 0.019) calculated from prediction mapping also revealed the best suitability of used bioclimatic variables in modelling (Supplementary Figure S1a,b). Further, the model output was also well supported with the calculated classification accuracy measures, such as K (0.391), NMI (0.611), and TSS (0.763) (

Table 1. Statistical measures of model performance and accuracy.

Measure	Calculated Value	Range Alongside a Description	Interpretation
AUC	0.975 ± 0.019	0–1; AUC ≥ 0.9 indicates very excellent; 0.9 > AUC ≤ 0.8 indicates good; AUC < 0.8 indicates poor [22,23,24].	Excellent
Kappa (K)	0.391	−1 to +1; K < 0.20, fair agreement; K = 0.21–0.40, moderate agreement; K = 0.41–0.60, good agreement; K = 0.61–0.80, substantial agreement; K ≥ 0.0.81, excellent/perfect agreement [28,29].	Near to good agreement
Normalized Mutual Information (NMI) n(s)	0.611	0 to 1; A value of 0 signifies complete inaccuracy of the models, whereas a value of 1 indicates precise prediction of the presence–absence relationship. [30]	Good prediction
True Skill Statistic (TSS)	0.763	−1 to +1, where “+1” denotes perfect agreement and “0” or lower values imply performance that is no better than random. [26]	Performance is better than the random model

), and other parameters shown in Supplementary Table S3. Based on the relative importance, the percentage contribution and permutation of each variable was assessed (

Table 2. The percentage contribution of environmental factors.

Label	Variable	Scaling Factor	Units	Percent Contribution	Permutation Importance
Bio 1	Annual Mean Temperature	10	°C	-	-
Bio 2	Annual Mean Diurnal Range	10	°C	5.9	1.7
Bio 3	Iso-thermality [(Bio 2/Bio 7) × 100]	100	°C	5.1	2.1
Bio 4	Temperature Seasonality (Std. Deviation × 100)	100	C of V	2.9	0.6
Bio 5	Max. Temperature of Warmest Month	10	°C	-	-
Bio 6	Min. Temperature of Coldest Month	10	°C	-	-
Bio 7	Annual Temperature Range (Bio 5–Bio 6)	10	°C	12.8	13.4
Bio 8	Mean Temperature of Wettest Quarter	10	°C
Bio 9	Mean Temperature of Driest Quarter	10	°C	2.0	14.3
Bio 10	Mean Temperature of Warmest Quarter	10	°C	-	-
Bio 11	Mean Temperature of Coldest Quarter	10	°C	-	-
Bio 12	Annual Precipitation	1	mm	-	-
Bio 13	Precipitation of Wettest Month	1	mm	3.2	11.3
Bio 14	Precipitation of Driest Month	1	mm
Bio 15	Precipitation Seasonality	100	C of V	29.5	10.3
Bio 16	Precipitation of Wettest Quarter	1	mm	0.2	1.5
Bio 17	Precipitation of Driest Quarter	1	mm	-	-
Bio 18	Precipitation of Warmest Quarter	1	mm	-	-
Bio 19	Precipitation of Coldest Quarter	1	mm	-	-
Alt	Altitude		m	9.8	14.4
Slop	Slope		°	3.0	0.6
Asp	Aspect		°	1.9	0.3
DNI	Direct Normal irradiance			3.6	6.7
Prec	Precipitation		mm	20.1	22.8
Tav	Average temperature		°C	-	-
Tmax	Maximum temperature		°C	-	-
Tmin	Minimum temperature		°C	-	-
Vap	Vapor		mm	-	-
Wind	Wind		m/s	-	-

Note: The MaxEnt program used a multi-collinearity test to select the highlighted (bold) bioclimatic variables for potential suitability modeling.

and Figure 1), and the variables, namely, precipitation seasonality (Bio 15), precipitation (Prec), annual temperature range (Bio 7), and altitude (Alt), showed highest level of percentage contribution (72.2%) and permutation importance (60.9%) for predicting the habit suitability of Th. spathiflorus in the northwestern Himalayas. Other variables like annual mean diurnal range (Bio 2), iso-thermality (Bio 3), and precipitation of wettest month (Bio 13) also showed important role in modelling; thus, important for predicting species ecological niche. The jackknife test (Supplementary Figure S2) indicated alt (Alt) as an important variable with maximum gain when used in isolation. Additionally, aspect (Asp) and slope (Slop) were two other important factors that showed significant decrease in gain when excluded.

The Bio 15 variable measures the variation in monthly precipitation over the course of the year, with a maximum probability (p) of presence at 60 (p = 0.85), 88 (p = 0.83), and a sharp decline at 73 (p = 0.28), as indicated by the response curves. Additionally, the precipitation variable exhibited the highest permutational importance (22.8) and the highest probability (p) of presence at 115 mm (p = 0.94) and 168 mm (p = 0.90), with a steep decline at 145 mm (p = 0.10). Bio 7 indicates that extreme temperature fluctuations did not significantly impact the distribution of Th. spathiflorus in the case of temperature variables. The habitat suitability range is 22 to 26 °C, with the highest probability (p) of occurrence at 23.0 °C (p = 1.00). Other factors, such as Alt (range = 2000–3500 m; maximum p = 0.88 at 2850 m) and DNI (maximum p = 0.85 at 17,200 Wm⁻²) also revealed the suitability of these parameters, which might affect the distribution pattern for effective prediction mapping of Th. spathiflorus.

3.2. Eco-Distribution, Shift Change, and KGCC Mapping of Th. spathiflorus

A total of 14 sites distributed in nine districts of two Himalayan states were surveyed, and majorly of occurrence points were collected from district Pithoragarh, followed by Shimla, Chamoli, and Uttarkashi (

Table 3. Geographic details of the surveyed sites, along with their associated species and status.

Sr. No.	Districts	Geospatial Data (Latitude/Longitude/Altitude, m)	Geo-Coordinates Recorded Areas	Associated Species	Population Distribution	Population Status
State: Himachal Pradesh (HP)
1.	Kullu	31⁰52′47.9″ 77°03′03.4″	Kokhan Wildlife Sanctuary	Acer spp., C. jacquemontii, Myrica esculenta, Pinus wallichiana, Quercus spp., Rhododendron arboreum	Random	Healthy
2.	Shimla	31°15.044′ 077°29.569′	Hatu	Abies pindrow, Acer spp., Corylus jacquemontii, Junglans sp., M. esculenta, Picea smithiana, P. wallichiana, Quercus spp., R. arboreum, Taxus wallichiana	Random	Healthy
		31°18.645′ 77°42.845′	Rampur		Random	Healthy
State: Uttarakhand (UK)
1.	Bageshwar	30°12′26.5″ 79°55′21.0″	Sunderdhunga	Trees: Q. foribunda, Q. semecarpifolia, R. arboreum, R. barbatum, Bamboos: H. falconeri	Uniform	Healthy
2.	Chamoli	30°41′21.41″ 79°35′22.51″	Ghangaria	Trees: A. pindrow, Acer spp., Betula alnoides, C. jacquemontii, Q. foribunda, Q. semecarpifolia, R. arboreum, R. barbatum, R. campanulatum, T. wallichiana, Bamboos: Drepanostachyum falcatum, Himalayacalamus falconeri, Yushania anceps,	Random	Disturbed
		30°28′12.81″ 79°13′8.59″ 2801	Chopta-Tungnath		Uniform	Pristine
3.	Dehradun	31°3′ 36.9″ 77°56′ 22.2″ 2435	Morach, Chakrata	Trees: A. pindrow, Juglans sp., P. smithiana, Q. semecarpifolia, R. arboreum, T. wallichiana,	Random	Reduced
4.	Pithoragarh	29°59′01’6″ 80°38′55.80″ 3055	Narayan Ashram, Dharchula	Trees: A. pindrow, Acer spp., Q. semecarpifolia, R. arboreum, T. wallichiana, Tsuga dumosa Bamboos: H. falconeri, Y. anceps	Uniform	Healthy
		30°01′41.20″ 80°38′51.30″ 2896	Karandam Bugyal, Dharchula		Uniform	Pristine
		30°10′06.19″ 80°35′00.31″ 2733	Darma valley, Dharchulla		Uniform	Healthy
5.	Rudraprayag	30°37′31.5″ 78°56′25.4″ 3117	Triyuginaraya	A. pindrow, Acer spp., B. alnoides, B. wallichiana, Carpinus viminea, C. jacquemontii, Fraxinus micrantha, Q. foribunda, Q. semecarpifolia,	Random	Healthy
			Agustayamuni	R. arboreum, R. barbatum, T. wallichiana Bamboos: D. falcatum, H. falconeri	Random	Healthy
6.	Tehri	30°41′10.2″ 78°40′12.5″ 2985	Pinsw	A. pindrow, Acer spp., Q. semecarpifolia, R. arboreum, T. wallichiana Bamboos: H. falconeri, D. falcatum	Random	Reduced
7.	Uttarkashi	30°58′54.1″ 78°26′53.9″ 2803	Yamunotri	A. pindrow, Acer spp., B. alnoides, B. utilis, Q. semecarpifolia, R. arboreum, R. barbatum, T. wallichiana	Uniform	Healthy
		31°07′18.5″ 78°21′46.1″ 2736	Har-ki-Do	A. pindrow, Acer spp., B. alnoides, B. wallichiana, C. viminea, C. jacquemontii, F. micrantha, Q. foribunda, Q. semecarpifolia, R. arboreum, T. wallichiana Bamboos: H. falconeri	Uniform	Pristine

). Based on the counts of clumps per 100 m, nine sites were classified as healthy, two pristine, two reduced, and one (from Ghangaria) disturbed. Similarly, a ratio of 1:1 was noted between random and uniform distribution of Th. spathiflorus populations. Overall, the distribution was recorded from 29°58′ (Narayan Ashram, Pithoragarh, UK) to 31°53′ (Ropa, Kullu, HP) in the north, and 77°02′ (Ropa, Kullu, HP) to 80°39′ (Narayan Ashram, Pithoragarh, UK) in the south, with the slope ranging from 45 to 85°. Attitudinally, the species spotted from the elevation ranged from 1914 to 3330 m.

The eco-distribution map for the current distribution and the estimated areas (in km²) are shown in Figure 2 and

Table 4. Estimated area under Th. spathiflorus in northwestern Himalayas revealed through MaxEnt modelling.

Sr. No.	Districts	Geographical Area (km2)	Forest Cover (km2)	Estimated Area (km2)	Estimated Area % in Respect to Total Forest Cover	Estimated Area % in Respect to Total Geographical Area
State: Himachal Pradesh
1.	Bilaspur	1167	380.70	-	-	-
2.	Chamba	6522	2455.16	10.03	0.41	0.15
3.	Hamirpur	1118	354.90	-	-	-
4.	Kangra	5739	2354.19	-	-	-
5.	Kinnaur	6401	645.99	-	-	-
6.	Kullu	5503	1976.29	15.43	0.78	0.28
7.	Lahaul & Spiti	13,841	160.35	-	-	-
8.	Mandi	3950	1773.02	37.82	2.13	0.96
9.	Shimla	5131	2419.41	213.05	8.81	4.15
10.	Sirmaur	2825	1390.87	-	-	-
11.	Solan	1936	890.29	-	-	-
12.	Una	1540	632.35	-	-	-
Total		55,673	15,433.52	276.33	1.79	0.50
State: Uttarakhand
1.	Almora	3144	1718	-	-	-
2.	Bageshwar	2241	1261	34.33	2.72	1.53
3.	Chamoli	8030	2709	141.26	5.21	1.76
4.	Champawat	1766	1224	-	-	-
5.	Dehradun	3088	1605	-	-	-
6.	Haridwar	2360	588		-	-
7.	Nainital	4251	3048	-	-	-
8.	Pauri	5329	3394	-	-	-
9.	Pithoragarh	7090	2078	152.07	7.32	2.14
10.	Rudraprayag	1984	1141	96.72	8.48	4.88
11.	Tehri	3642	2065	16.08	0.78	0.44
12.	Udham Singh Nagar	2542	436	-	-	-
13.	Uttarkashi	8016	3028	265.00	8.75	3.31
Total		53,483	24,295	705.46	2.90	1.31
Grand total		109,156	39,728.52	981.79	2.47	0.90

, respectively. This study provides an actual occurrence area of ~136 km² and estimated area of ~982 km² under Th. spathiflorus in both the Himalayan states. Among the districts, maximum and minimum estimated areas were recorded in Uttarkashi (~265 km²) and Chamba (~10 km²), respectively. Notably, the total area occupancy of the species was calculated as 2.47% of the forest cover and 0.90% of the total geographical area (Supplementary Table S4). Additionally, the eco-distribution map was superimposed on ASTER GDEM to calculate the area cover in accordance with the altitudinal gradient. The maximum occupancy (~331 km²) of the species lies between 2501 and 3000 m amsl, showing a sharp decline below and above. Notably, above 3251 m, no record of Th. spathiflorus was observed in state HP but found ~119 km² area in UK (Supplementary Table S5).

The probabilistic distribution for current and future climatic scenario (as per RCP 8.5) of 2050 and 2070 are shown in Figure 3 and Supplementary Figure S3a,b. Model showed a sharp decline in area of an optimal habitat (~966 km² in 2050 and ~936 km² in 2070) for future RCP 8.5 scenario with respect to the current estimation, i.e., ~982 km². The districts Solan (HP), Uttarkashi (UK), and Pithoragarh (UK) showed a maximum climatic shift in current suitable habitats. Notably, the predicted suitable habitat was moved towards “East by North (EbN, 88°)” for year 2050 and an eastern shift (91°) by the year 2070 (

Table 5. Current distribution centroid point shifting analysis with RCP 8.5 of 2050 and 2070 for Th. spathiflorus in northwestern Himalayas, where EbN = east by north.

Sr. No.	Factors	Current	2050	2070
Himachal Pradesh
1	Long–Lat	77°25′01.78″ E 31°12′18.60″ N	77°27′34.60″ E 31°12′45.21″ N	77°28′01.70″ E 31°12′16.46″ N
2	Direction/degree (°)	0	EbN (83°)	East (90°)
3	Distance (km)	0	4.1	4.77
4	Area (km2)	276.33	262.37	246.23
Uttarakhand
1	Long–Lat	79°10′04.37″ E 30°36′40.67″ N	79°12′26.31″ E 30°36′56.71″ N	79°12′32.59″ E 30°36′52.71″ N
2	Direction/degree (°)	0	EbN (85°)	East (88°)
3	Distance (km)	0	3.78	3.97
4	Area (km2)	705.46	703.12	689.38
Combined for both the states
1	Long–Lat	78°43′19.28″ E 30°45′54.33″ N	78°47′26.38″ E 30°46′48.61″ N	78°47′58.24″ E 30°45′46.78″ N
2	Direction/degree (°)	0	EbN (88°)	East (91°)
3	Distance (Km)	0	5.74	6.4
4	Area (km2)	981.79	965.49	935.61

).

Additionally, Figure 4 depicts the probabilistic distribution of MIROC6-SSP245 for the current and future climatic scenarios of 2021–2040 and 2061–2080, respectively. The model, however, demonstrated that the area of an optimal habitat for the future MIROC6-SSP245 scenario decreased in comparison to the current estimation. Specifically, it decreased from ~982 km² (current prediction 1970–2000) to ~950 km² in 2021–2040, and then to ~928 km² in 2061–2080. Uttarkashi, Bageshwar, and Pithoragarh (UK), as well as the districts of Kullu, Kangra, Chamba, and Solan (HP), demonstrated the most substantial climatic shift in the habitats that are presently suitable. The relocation of the anticipated suitable habitat for the years 2021–2040 and 2061–2080 from a lower altitude to a higher altitude is crucial to observe.

Overlaying of current distribution of Th. spathiflorus over KGCC map revealed its occurrence in six climatic subtypes prevailed in the northwestern Himalayan states (Figure 5). Importantly, the maximum occurrence of species was found in the subtropical highland oceanic climate (Cwb; C is warm temperate, w is winter dry, and b is warm summer) of the middle Himalayas in the districts of Shimla, Uttarkashi, Rudraprayag, Bageshwar, Chamoli, and Pithoragarh. The distribution was followed by a monsoon-influenced warm summer humid continental climate (Dwb; D is snow, w is winter dry, and b is warm summer) and subarctic climate (Dwc; D is snow, w is winter dry, and c is cool summer). Unexpectedly, some traces of species occurrence were also observed in the humid subtropical climate (Cwa; C is warm temperate, w is winter dry, and a is hot summer) of lower stretches of Dhaola Dhar and Shiwalik ranges in HP and UK, respectively. Lastly, species recorded to occur in the glacial (ET; E is polar and T is polar tundra) climatic condition of the eastern regions where the temperature of warmest month varies from 0 to 10 °C.

4. Discussion

In India and around the world, bamboos play a critical role in the generation of livelihoods and the maintenance of wood-based industries from a socioeconomic perspective. Furthermore, they offer substantial ecological services, including the preservation of biological diversity, the prevention of soil erosion, and the conservation of soil moisture [31,32]. The last few decades have witnessed a significant change in the global climate of the Earth, particularly in the Himalayas, and therefore, it is important to investigate its impact over the range expansion or shrinkage of the vegetation, particularly climatic climax species of woody perennials [33] (http://www.ipcc.ch/, accessed on 16 February 2024). Th. spathiflorus is a temperate bamboo species occurring at subtropical to temperate zone of the Himalayas and occupying a highest altitudinal range up to ~3300 m between tree lines and alpine steppes. Also, it showed a wide geographical span in northwestern Himalayas ranging from 29°58′ to 31°53′ in north and 77°02′ to 80°39′ in east. A variety of dominant tree species, including Abies pindrow, Acer spp., Betula alnoides, Quercus foribunda, Q. semecarpifolia, Rhododendron arboreum, R. barbatum, Taxus wallichiana, and Tsuga dumosa, dominate the understory shrubby layer at an altitude of 2500 to 3200 m. The two temporal bamboo species, Yushania anceps and Himalayacalamus falconeri (commonly referred to as ringal in UK and nigal in HP), were also present (Table 3). At some sites, it was observed to be extended up to the tree line, where it was grown in association with the Abies spectabilis, Betula utilis, Rhododendron campanulatum, etc. All these physiographic features and geographical attributes revealed that it acts as an important link between different forest clines and play significant ecological functions towards sustaining biological diversity.

4.1. Model-Based Habitat Suitability and Associated Bioclimatic Variables

We also assessed MaxEnt using various statistical parameters, such as AUC, K, TSS, and NMI, to confirm its reliability and statistical support. Model output with a high AUC value close to 1 (0.975 ± 0.019) and good scores on other classification accuracy tests show that the bioclimatic variables used to run the model were chosen correctly to make a good prediction (Table 1 and Supplementary Figure S1), which was found in accordance with earlier studies by [26,28,29]. Importantly, while working on such models’ spatial scale, selection for niche prediction are also depend upon the size and areal extent of the distribution range, which should be precisely chosen and surveyed thoroughly [34]. Thus, the reason for the extraordinarily high accuracy of MaxEnt achieved in this study could be attributed to the extensive coverage of distribution range, sampling accuracy, and site phonographic features recorded during the surveyed period. Earlier projection assessments in bamboo revealed the prediction with AUC value 0.914, 0.929, and 0.933 for presence-only data, presence and true-absence data, and presence and pseudo-absence data, respectively [35]. However, an ensembling approach and conglomeration of modelling tools has been applied for difficult mountain terrain for projecting bamboo species, where baseline climate dataset was used for model calibration. Thus, enforcing gridded datasets from observatories and satellite measurements estimated the uncertainty in species distribution exercise [36].

Further, response curves, percentage contribution and permutation importance, and the jackknife test were also used to reveal the importance of the predictors used in this study. Effective variables included Bio 15 (maximum probability of presence at 60 (p = 0.85) and 88 (p = 0.83), Prec (p = 0.94 at 115 mm and p = 0.90 at 168 mm), Bio 7 (p = 1.00 at 23.0 °C), Alt (p = 0.88 at 2850 m), Bio 2, Bio 3, Bio 13, and DNI (p = 0.85 at 17,200 Wm⁻²) for predicting the Th. spathiflorus distribution range in northwestern Himalayas. In comparison to Th. spathiflorus, the response curves of Oxytenanthera abyssinica demonstrated that the most significant factors defining the habitat suitability range are the precipitation of the coldest quarter (Bio 19), the precipitation of the warmest quarter (Bio 18), iso-thermality (Bio 3), the precipitation of the driest quarter (Bio 17), slope, and precipitation seasonality (Bio 15) [37]. It suggests that the precipitation fluctuation plays a crucial role in predicting the habitat suitability of high-altitude bamboos. In another ENM-based study [38], temperature was suggested an important variable influencing horizontal and vertical expansion of invasive bamboo species of Japan, such as Phyllostachys edulis and P. bambusoides. An earlier study [39] used a combined modeling approach to employ a distribution model of species, with bamboo suitability as one of the factors, along with other bioclimatic variables (AUC = 0.932). The bamboo suitability model consistently does a better job than the bioclimatic and combined models at predicting the growth of Ailuropoda melanoleuca (Giant Panda) in the Qinling Mountains (China). Similar ENM studies were also performed in other plant species like S. purpurea in the Tibetan plateau [40] and Rhododendron ponticum in the United Kingdom [41]. These studies signify that the temperature and rainfall extremes are the key climatic regimes of the mountain region and had a strong influence over the suitability of vegetation as well as wildlife there.

4.2. Eco-Distribution Mapping and Estimation of Area Cover under Th. spathiflorus

The eco-distribution mapping revealed probabilistic distribution of species in both the states of northwestern Himalayas with estimated area occupancy of ~982 km². Across the districts, maximum area under species was recorded in Uttarkashi (UK; ~265 km²) and minimum for Chamba (HP; ~10 km²). Altitudinally, maximum probability of species distribution was recorded between the altitudinal range of 2501 and 3000 m, which sharply declined above and below (Table 4). In Nepal’s Himalayas, a study concluded that the strongest predictors of red panda distribution were tree and bamboo cover, proximity to water bodies, and aspect. The presence of bamboos was observed in 85% of sign plots, as the majority of hilly bamboos are often associated with Ailurus fulgens (Red Panda) [42]. The genetic database and ENM tools were recently employed to investigate two bamboo partridges (Chinese and Taiwanese). The results indicated that the habitats were highly conserved and shared an overlapping distribution range during the evolutionary time period [43]. Therefore, the significant increase in bamboo-covered land may correlate with the distribution of faunal life, and the opposite may also hold true. The ENM has been widely used recently for predicting habitat suitability and climate change scenarios in Himalayan species, such as Guadua angustifolia [44], Myrica esculenta [45], Quercus lanata [46], Quercus semecarpifolia [47], Rhododendron arboreum [48], etc.

The current study revealed a strident decline of 1.66% and 4.70% for future RCP 8.5 scenario during 2050 and 2070, respectively. Importantly, an overall eastern by northward shift has been predicted for centroid location of distribution. Contrarily to our findings, the back-propagation neural network (BPNN), Markov chain, and cellular automata (CA) coupling model showed an expansion trend for bamboo forest, especially in the cultivated land of Anji County (Zhejiang Province, China) under RCP 2.6, 4.5, 6.0, and 8.5 [49]. Overall, in China, potential distribution of bamboo showed an expansion of ~91,500 km² from 1961 to 2099 by using a support vector machine (SVM) model for climate change scenarios [35]. In a study conducted on an Ethiopian bamboo Oxytenanthera abyssinica, it was observed that the total area of high-potential regions will increase and least-potential regions would decrease under the future climate change scenarios of three RCPs (RCP2.6, RCP4.5, and RCP8.5) during the 2050s and 2070s [37]. These data suggested that high-altitude alpine bamboo Th. spathiflorus may be considered as a climatic indicator among the herbaceous species. For a better understanding of the potential habitat suitability distribution of Th. spathiflorus, this current investigation utilized MIROC6-SSP245 for the scenarios 2021–2040 and 2061–2080, according to the future RCP 8.5 scenario outputs for 2050 and 2070, which declined due to climate shift as well as the habitat suitability zones to ~950 km² in 2021–2040 and then to ~928 km² in 2061–2080.

Further, overlaying of estimated distribution over KGCC allowed for delineating the area occupancy divided into six major climatic classes. The KGCC aggregates complex climate gradients into a simpler one by accounting of ecologically meaningful schemes [50], and is used to analyze the distribution [51,52,53], growth behavior of species [54], and setting-up of dynamic global vegetation models [55]. It was observed that the highest estimated area cover was demarcated in “Cwb” zone, which has been characterized as a highland type of climate with temperature of hottest month < 22 °C, cold and dry winters, and warm and wet summer. These climatic scenarios prevailed in the upper ridges of districts Shimla, Uttarkashi, Rudraprayag, Bageshwar, Chamoli, and Pithoragarh in the middle Himalayas, where the winters are long, dry, and severely cold while, the summers are short and cold. Likewise, subarctic climate (Dwc) was predominated in Pithoragarh where winters are long and cold with short and mild summer. Furthermore, tundra-type climate (ET) also occurs in Karanadam Bugyal (alpine meadow) in Pithoragarh, where summers are very short and cold with the temperature of warmest month lying between 0 and 10 °C. Herein, the effective use of KGCC was recognized for Th. spathiflorus distribution in northwestern Himalayas, which tends to reflect those climatic data, being the major driver of global vegetation distribution [30,56,57].

5. Limitations of the Model and Study

The MaxEnt model protocol handles massive number of datasets at once, which might cause data handling, generalization, and interpretation concerns. It also requires a lot of storage and fast computers. Two protected regions and challenging terrain states (HP and UK) provided data for this investigation. Ground surveys and data collection require permission, which were generally tough across the terrain. Therefore, we skipped several sites in the Indian and Nepalese Himalayas. It requires ground-based occurrence data and secondary data on forest types, forest cover, soil, climate, and environmental variables, all of which have varying spatial resolution. The number of variables boosts model accuracy, but too much spatial heterogeneity can mislead the model. The processes of collecting field data through sample and survey, acquiring RS and agency-specific categorical data, and other related tasks are time-consuming, costly, and necessitate the involvement of expert individuals. When there are insufficient species data to generate precise maps or extrapolations of species distribution, experts design SDMs. However, theoretical issues and poor research have reduced SDM outputs, decision-making, and environmental licensing. It is challenging to compare MaxEnt results with other algorithms because they provide environmental appropriateness and suitability for bamboos and other grass species, which are located beneath the canopy, thereby reducing the projected likelihood of occurrence. Instead of comprehensive field surveys and estimations, MaxEnt’s logistic output is based on the prevalence assumption for environmental appropriateness.

6. Conclusions

This study presents a high-confidence potential distribution map of Th. spathiflorus in northwestern Himalayas for the present and future climatic scenario and provides an estimation of the area cover under the species. This is pioneering information generated for the mapping and area cover of this high-altitude bamboo species from the Himalayas, which could be wisely utilized by the forest managers and conservationists. This study provided a fair idea of the important bioclimatic variables associated with the habitat suitability of this species, which is immensely important to understanding the ecological niche of the high-altitude bamboos of the Himalayas. Lastly, future climate change scenarios indicated a habitat shrinkage and range shift (eastern by northward) in the investigated bamboo species.